Takanori Yamashita

High-Quality Telemedicine Using Digital Video Transport System over Global Research and Education Network

Shuji Shimizu1, Koji Okamura2, Naoki Nakashima3, Yasuichi Kitamura4, Nobuhiro Torata5, Yasuaki Antoku6, Takanori Yamashita7, Toshitaka Yamanokuchi8, Shinya Kuwahara9 and Masao Tanaka10 1,2,3,5,6,7,8,9,10Telemedicine Development Center of Asia, Kyushu University Hospital, 1Department of Endoscopic Diagnostics and Therapeutics, Kyushu University Hospital 2Research Institute for Information Technology, Kyushu University 3,6,7,8Department of Medical Informatics, Kyushu University Hospital 4National Institute of Information and Communications Technology 9Kyushu Electric Power Company 10Department of Surgery and Oncology, Kyushu University Graduate School of Medical Sciences Japan

Standard measure and SVM measure for feature selection and their performance effect for text classification

This paper compares the prediction performance of document classification based on a variety of feature selection measures. Empirical experiments were conducted for the dataset re0 with 10 measures for feature selection and with SVM. It is confirmed that the feature selection based on the SVM-score proposed by Sakai and Hirokawa (2012) outperforms the standard measures with small number of features. In fact, 100 words are enough to get the similar performance obtained with all words. The reason of good performance of this feature selection is that the SVM-score capture not only the characteristic words of positive samples but of negative samples as well.

Classification of Imbalanced Documents by Feature Selection

We previously worked on category classification problem of reuter s newspaper article using SVM and feature selection. In the study, feature selection by SVM-score [Sakai, Hirokawa, 2012] showed high accuracy. It was also expected to be superior to other standard indicators in case data is imbalanced. This study aimed to show the effectiveness of feature selection by SVM-score in machine learning with imbalanced data. For the reuters data, F-measure was calculated in the classification experiment of all 13 categories. As a result, feature selection by SVM-score shows high f-measure and precision. In addition, we found feature words of negative example improve the classification performance.

Classification and Feature Extraction for Text-based Drug Incident Report

Medical institutions have been constructed incident report system, then accumulating incident data. Incident data compose text-based data and some structured attributes. We considered based on the analysis result with clustering for drug incident report. Firstly, we generated a network of documents and words from the text-based data. Secondly, Louvain method was applied to the network and 11 clusters were generated. We confirmed the contents of each cluster from feature words extracted by TF-IDF. Then, we compare clusters of text-based data with structured attributes and grasp the trend of the incident. This proposed method showed the possibility of clinical support toward reduction incident from text-based data.

Temporal Relation Extraction in Outcome Variances of Clinical Pathways.

Recently the clinical pathway has progressed with digitalization and the analysis of activity. There are many previous studies on the clinical pathway but not many feed directly into medical practice. We constructed a mind map system that applies the spanning tree. This system can visualize temporal relations in outcome variances, and indicate outcomes that affect long-term hospitalization.

Visualization of Key Factor Relation in Clinical Pathway

Abstract The secondary use of medical data to improve medical care is gaining much attention. We have analyzed electronic clinical pathways for improving the medical process. The analysis of clinical pathways so far has used statistics analysis models, however as issue remains that the order, and multistory spatial and time relations of the each factor could not be analyzed. We constructed an Outcome tree system that shows the greatest significant relation for each factor. The Hip replacement arthroplasty clinical pathway was analyzed by the system, and the outcome variance of the clinical pathway was visualized. The results indicate the path of patients who have a long hospitalization stay and extracted four critical indicators.

Construction of Dominant Factor Presumption Model for Postoperative Hospital Days from Operation Records

The secondary use of clinical text data to improve the quality and the efficiency of medical care is gaining much attention. However, there are few previous researches that have given feedback to clinical situations. The present paper analyzes the words that appear in operation records to predict the postoperative length of stay. SVM (support vector machine) and feature selection are applied to predict if a stay is longer than the standard length of 25 days. It was confirmed that with less than 20 feature words we can predict if a stay is longer or not with almost the optimal prediction performance.

Extraction of determinants of postoperative length of stay from operation records

Secondary use of clinical text data are gaining much attention in improving the quality and the efficiency of medical treatment. Although there is some case studies of medical-examination text data, there are not many examples fed back to the medical-examination spot. The present paper analyses the operation records of total hip arthroplasty. We extracted feature words that characterize the two peaks which appeared in distribution of postoperative hospital days using SVM (support vector machine) and FS (feature selection). The models gained by optimal FS attained 60% accuracy as prediction performance. We applied logistic regression analysis to estimate postoperative length of stay from the extracted feature words. Most words were not statistically significant except two words.